Reliability of mySugr hba1c estimation

Status
Not open for further replies.
M

Mgfg1rl

Member
Relationship to Diabetes
Type 2
Hi does anyone use mySugr? I wonder how reliable the estimated hba1c is? Any thoughts welcome . Thank you
 
It will very much depend upon when you test and how often you test as to how reliable the prediction will be, but I would say probably somewhere in the +/- 5-10mmol/mol range. If you only ever test after food it will likely estimate slightly higher than your actual reading, if you test equally before and after food then you will get a better balance of results. If you only or mostly test before meals, then it will likely underestimate your HbA1c. The more frequently you test the more accurate the prediction should be.... but it is still only a prediction. Unless you set an alarm and test through the night as well as during the day, you are missing a big chunk of data regarding what is happening with your levels and they rarely stay level overnight

Libre which samples our glucose levels about every 5 mins 24/7 usually estimates a few mmols lower than actual HbA1c, but then I find Libre usually reads about 1mmol lower than actual finger prick BG. The point is that it averages a lot more readings than maybe just 4 or 5 per day.... or perhaps just 1 or 2 a day depending upon how frequently or infrequently you test and how representative that is of your BG fluctuation, which is what a My Sugr app has to work off.
 
I personal find it and my Libre estimation both are lower than the actual laboratory one.
 
I used my my suger I found that last time my HB1AC was 5 mmolls below what mySuger was was estimateing for that time period but I think it will different for different people depending on variety of stuff.
 
I think it will vary from person to person.

Generally, like @grovesy I’ve always found the predictions of various formulae to be a little optimistically low (including from continuous sensors that should have all the data to work with).

There is no mathematical way to convert between average glucose and HbA1c as the two are measuring very different things - so the HbA1c suggestions can only ever be a vague guide.
 
If any of you Libre users are either sufficiently nerdy or sufficiently interested then you need to do what I would do. That is to collect the data and record the libre prediction at the time blood was taken for laboratory measurement. You could then get a correlation between the two which you could then use for predictive purposes. Would take a few years to get the numbers so, as I say, you would need to be sufficiently nerdy or sufficiently interested.

Alternatively, if any of you have bits of data and are happy to publish libre predictions at the time blood was taken and the subsequent Hba1c result, I would be more than happy to look at the data as a block and try and answer this question a bit more scientifically.
 
Docb said:
collect the data and record the libre prediction at the time blood was taken for laboratory measurement. You could then get a correlation between the two which you could then use for predictive purposes.
Click to expand...
I have a spreadsheet so I can track how my HbA1c has changed over the years. I also note the HbA1c the Libre app says on the day of the test, but as others have noted, the Libre app usually predicts a bit lower than what the blood test says.

I also have two columns on the sheet where on the day of my blood test, my partner and I can guess what it will come back as.
 
Interesting @Finn. If you plot predicted against actual do you get any sort of relationship?
 
Finn said:
I also have two columns on the sheet where on the day of my blood test, my partner and I can guess what it will come back as.
Click to expand...

Hehe! We used to have occasional threads where forum members would post their previous result and a brief summary of how things had been going recently on the day they had their bloods drawn, giving forum members a few days to offer their guesses as to what the new result might be.

Not seen one of those for a little while! Like this…

 
Docb said:
If you plot predicted against actual do you get any sort of relationship?
Click to expand...
I haven't been tracking the Libre app's prediction long enough to see a pattern, but based on the little data I have, it seems like my actual HbA1c tends to be about .3 higher than the Libre app's prediction.

My Libre app currently predicts 5.6% (with only 1% of time spent in the red in the last 90 days! ). I have a test next week and I'll guess that my actual HbA1c will be closer to 6.0%. I think anything under 6% would be a bit magical, but I'd certainly take it!
 
Finn said:
This needs to be started back up!
Click to expand...

I was thinking just the same! I’ll try to remember when I get my next HbA1c check - which is probably February
 
Docb said:
If any of you Libre users are either sufficiently nerdy or sufficiently interested then you need to do what I would do. That is to collect the data and record the libre prediction at the time blood was taken for laboratory measurement. You could then get a correlation between the two which you could then use for predictive purposes. Would take a few years to get the numbers so, as I say, you would need to be sufficiently nerdy or sufficiently interested.

Alternatively, if any of you have bits of data and are happy to publish libre predictions at the time blood was taken and the subsequent Hba1c result, I would be more than happy to look at the data as a block and try and answer this question a bit more scientifically.
Click to expand...
As regards using Libre data to predict HbA1c, personally I think this will always be a non-starter because there is just too much variability between sensors. I ran an analysis some time ago after having had a few sensors where I had to calibrate more than usual with xDrip+. The difference between blood glucose as reported by Libreview versus what I had logged as the xDrip+ calibrated value was quite significant, especially for that last 2-3 days of one sensor in particular. See attached excerpt from a PDF I wrote up for my own use at the time.
I think that even trying to predict HbA1c based on finger-prick (and calibrated CGM) blood-glucose values is never going to be accurate because [1] there is variability between accuracy of glucose meters and boxes of test strips and [2] as @everydayupsanddowns points out, there is no direct formula that will give a HbA1c (lab accurate) number based purely on average blood glucose.
As I understand it, HbA1c is a function of how long the patient exists in a state where his/her blood glucose (as measured in a lab environment) is elevated above a defined value (from memory, I think the trigger value is 8 mmol/L, but I could be wrong). So, one person could have a lower HbA1c than another even though the former had poorer control (e.g., more hypos) than the latter.
Personally, I focus on TIR and the shape of the AGP plot. Basically, I try to keep between the target min/max values as much as possible and try to react as quickly as I can to Libre-supplied trends so as to minimise peaks and troughs.
As ever, I hope all that makes sense?
 

Attachments

If you have Libre data, why care much about HbA1c? It's only purpose is to give a rough estimate of a summary of some of the kind of data you get from a Libre.
 
Well, I care about HbA1c because my diabetes care team do, firstly, and because it measures medical markers that are left in my body purely as a result of my having diabetes. The point I was trying to make in my earlier post is that HbA1c is difficult, if not impossible, to accurately predict purely based on blood glucose. My own predictions are pretty close, but never match what the lab results are. I, however, have stopped worrying about the mismatch.
 
littlevoice359 said:
Well, I care about HbA1c because my diabetes care team do, firstly, and because it measures medical markers that are left in my body purely as a result of my having diabetes. The point I was trying to make in my earlier post is that HbA1c is difficult, if not impossible, to accurately predict purely based on blood glucose. My own predictions are pretty close, but never match what the lab results are. I, however, have stopped worrying about the mismatch.
Click to expand...
The fundamental reason why you can't calculate HbA1c from BG (or avg BG from HbA1c) is because the HbA1c corresponding to a particular avg BG is different for everybody. The variability comes mainly from variability between people in the avg lifetime of red blood cells (which can itself change over time for any individual). Having perfectly accurate sensors wouldn't change the picture.

HbA1c is really just a clever kludge.
 
littlevoice359 said:
As regards using Libre data to predict HbA1c, personally I think this will always be a non-starter because there is just too much variability between sensors. I ran an analysis some time ago after having had a few sensors where I had to calibrate more than usual with xDrip+. The difference between blood glucose as reported by Libreview versus what I had logged as the xDrip+ calibrated value was quite significant, especially for that last 2-3 days of one sensor in particular. See attached excerpt from a PDF I wrote up for my own use at the time.
I think that even trying to predict HbA1c based on finger-prick (and calibrated CGM) blood-glucose values is never going to be accurate because [1] there is variability between accuracy of glucose meters and boxes of test strips and [2] as
Click to expand...
Totally agree. Both are inaccurate, especially sensors. The ones I've used became so inaccurate towards the end I stopped scanning them.

littlevoice359 said:
As I understand it, HbA1c is a function of how long the patient exists in a state where his/her blood glucose (as measured in a lab environment) is elevated above a defined value (from memory, I think the trigger value is 8 mmol/L, but I could be wrong). So, one person could have a lower HbA1c than another even though the former had poorer control (e.g., more hypos) than the latter.
Click to expand...

It's a measure of how many blood cells have glucose attached to them over a three month period. Normal levels are between 4% and 6% of blood cells. If blood sugar has been raised for long periods of time, more cells will have glucose attached to them, hence rises that do not last long shouldn't impact the value as fewer cells will be affected.
 
Eddy Edson said:
The fundamental reason why you can't calculate HbA1c from BG (or avg BG from HbA1c) is because the HbA1c corresponding to a particular avg BG is different for everybody. The variability comes mainly from variability in the avg lifetime of red blood cells (which can itself change over time for any individual). Having perfectly accurate sensors wouldn't change the picture.

HbA1c is really just a clever kludge.
Click to expand...

And for some people hba1c doesn't actually work - there's something in blood that can make the reading totally wrong. (Can't remember what it is!)
 
harbottle said:
And for some people hba1c doesn't actually work - there's something in blood that can make the reading totally wrong. (Can't remember what it is!)
Click to expand...
Different kinds of anemia can screw it up in different ways. And there's other rarer stuff too I think.
 
Status
Not open for further replies.
Back
Top